This blog post argues that achieving reliable long-horizon autonomy in coding agents is unlikely, highlighting the gap between claims of success in the field and the actual challenges of long-term operation. The author provides insights from industry data to support this view, making a case for skepticism toward those who assert otherwise.